Research Select Project

EpiBench

Benchmarking suite for DNA methylation detection methods. CNN/PyTorch framework for methylation classification from bisulfite sequencing data, benchmarked against traditional statistical approaches.

Source ↗
Bioinformatics PyTorch Machine Learning Genomics

EpiBench is a benchmarking framework for evaluating DNA methylation detection methods across bisulfite sequencing datasets. The core is a convolutional neural network built in PyTorch, trained to classify differentially methylated regions from WGBS data.

Motivation

Traditional methylation callers (Bismark, MethylDackel) produce call sets, but comparative evaluation across methods on matched datasets is sparse. EpiBench provides a reproducible harness for head-to-head benchmarking — same data, same evaluation criteria, different detection approaches.

Technical Approach

The CNN architecture processes fixed-length sequence windows centered on CpG sites, incorporating methylation fraction, coverage depth, and local sequence context as input features. Performance is evaluated against DSS and BSmooth as baseline statistical comparators.

Status

Research prototype. In active development.